GGUF quants of Dispatch-7B, an agent orchestrator built on Mistral 7B that routes agentic work: request + tool catalog in → JSON execution plan out (which tool, what order, when to escalate to a frontier model). 97.4% valid-plan rate, ~87% estimated cost reduction vs. frontier-does-everything — full evaluation and usage details on the main model card.

File	Quant	Size	Use
`mistral-7b-instruct-v0.3.Q4_K_M.gguf`	Q4_K_M	4.4 GB	recommended — fits 8 GB RAM/VRAM
`mistral-7b-instruct-v0.3.Q8_0.gguf`	Q8_0	7.7 GB	near-lossless local quality

Run with Ollama

ollama run hf.co/hvss/Dispatch-7B-GGUF:Q4_K_M

Run with llama.cpp

llama-cli -m mistral-7b-instruct-v0.3.Q4_K_M.gguf --temp 0 -p "[INST] <dispatcher system prompt + tool catalog>

<user request> [/INST]"

Use the exact dispatcher system prompt from the main model card, with your own tool catalog substituted in. Greedy decoding (--temp 0) recommended — plans are structured output, not creative writing.

License: Apache 2.0.